This weekly update brings big news about AI coding agents and how they're really doing in the real world.

The biggest story came from a bold prediction that didn't come true. Six months ago, Anthropic's CEO said that by September 2025, AI would write 90% of all code. Well, September 2025 is here, and that prediction was completely wrong. New research shows that AI coding tools are actually making developers slower, not faster. Developers spend less time writing code, but they waste more time fixing AI mistakes and waiting for the systems to work.

Meanwhile, OpenAI quietly tested new AI agent features in ChatGPT. Some users spotted mysterious "alpha models" that could act like smart agents. The models had names like "Agent with truncation" and "Agent with prompt expansion," but OpenAI quickly removed them after people noticed.

Meta released a new AI model called CWM that's designed specifically for coding and understanding how the world works. This 32-billion parameter model can help with code generation and reasoning tasks.

There's also good news for developers working with huge codebases. A new technique called "frequent intention compaction" helps AI agents work better with large, complex code projects. One team used this method to handle Rust code with 300,000 lines and finish a week's worth of work in just one day.

For people who want to build their own AI agents, there's now a 15-minute tutorial for creating a YouTube research agent using Claude Code. This shows how easy it's becoming to build useful AI tools.

The reality check is clear: while AI coding tools are getting more powerful, they're not the magic solution many companies hoped for. The real challenge is learning how to use them properly and fixing the problems they create.